Skip to main content
Share of Voice (SoV) measures how much of the answer you own. Every prompt is sent to each engine repeatedly, and each answer is scored on its own: your share is your brand set against every other brand mentioned in that same answer. Share of Voice is a share of the airtime, so how often you are named inside an answer matters. A brand the engine returns to again and again owns more of that answer than one it lists in passing, and your share reflects that. Answers where you were not mentioned at all still count, as a zero. That makes SoV a combined signal: it falls when you appear less often, and it falls when you appear alongside more competitors. To separate those two causes, zued also reports Share when present, which averages only the answers you actually appeared in.
Share when present is the one to read next to Appearance Rate, because the two separate the causes: Appearance Rate tells you how reliably you make it into the answer, Share when present tells you how dominant you are once you’re in it. If Share of Voice falls, those two say which of the two things happened. They are two views of the same underlying answers, not two factors to combine. Reading them together tells you where to act; arithmetic between them does not produce a third number worth having. Brand mentions are extracted and normalized from every AI response collected in your project. Normalization handles variations like abbreviations and alternate spellings so they all count toward the same brand. By default a brand’s figure also includes its products and services, so a brand and the things it sells are read as one; the specific-name view shows each name on its own. By default these numbers are measured on unbranded prompts: questions that do not name your brand. A prompt that names it is answered with your brand in it almost by construction, so it would lift the figure without saying anything about the questions a stranger asks. zued sets those prompts aside and shows them under the Branded filter, on their own or together with the rest. What counts as your brand follows the view you are in: at the brand level your products and, where you count it, your network belong to it; in the specific-name view only your tracked name and its spellings do. Only brands the AI recommends, compares, or discusses in its own prose are counted. Brand names that appear only inside citation links, source attributions, or the source, product and map cards an engine displays are not counted as mentions. See Mentions, Citations & Sources for the full distinction. Only names that are brands enter the share. Names the engines use for a kind of thing, a place or a law are kept out, and so is anything you switched off or set to another kind on the Brands page. Your tracked brands always count. Every Share of Voice figure in zued, on the dashboard, per page and in the competitive landscape, counts the same names.

​
What it tells you

Both numbers measure dominance: when the AI brings you up, is the answer about you, or are you one name in a long list? SoV folds in how often that happens at all, Share when present holds it constant. The pairing with Appearance Rate is what makes it actionable. High Share when present with low Appearance Rate means you dominate the few answers you reach, so the bottleneck is getting into more answers. High Appearance Rate with low Share when present means you’re reliably present but diluted, so the lever is stronger positioning and depth. The competitive landscape plots exactly these two axes for every brand the engines mention. SoV remains a secondary metric. A high SoV with a low Alignment Score means you’re being mentioned but possibly misrepresented. Address alignment first, then use SoV to track competitive position over time.

​
Per-engine breakdown

SoV is tracked per engine. Your aggregate SoV can look healthy while you’re absent from a specific engine entirely. Check per-engine breakdowns to catch engine-specific blind spots.
Since zued audits weekly, SoV is a directional indicator, not a real-time measurement. It’s reliable enough to spot trends and compare engines, but take individual data points with a grain of salt.